In [1]:
from flytekit.core.context_manager import FlyteContextManager
In [2]:
path = None
In [3]:
# Parameters
path = "/var/folders/jx/54tww2ls58n8qtlp9k31nbd80000gp/T/flyte-ytmvsbhm/raw/a0499d0fbc48ab90edf515f297d056ea"
In [4]:
from IPython.display import display
import pandas as pd
from PIL import Image

Metrics Report¶

In [5]:
df = pd.read_parquet(path)
df["correct"] = ((df.prediction > 0.5) == df.label).astype(int)
df.sort_values(by="correct")[["label", "prediction", "description"]]
Out[5]:
label prediction description
2 0.0 0.854929 This image is named img3.png and contains clouds
0 1.0 0.963508 This image is named img1.png and contains hydr...
1 1.0 0.960557 This image is named img2.png and contains hydr...
In [6]:
acc = df["correct"].sum() / len(df)
print(f"accuracy = {acc}")

accuracy = 0.6666666666666666

False Positives¶

In [7]:
ctx = FlyteContextManager.current_context()
In [8]:
for i, img in df.iterrows():
    if (img.prediction > 0.5) == 1 and img.label == 0.0:
        print(f"Image {i}: {img.description}")
        local_img = ctx.file_access.get_random_local_path()
        ctx.file_access.get(img.remote_source, local_img)
        display(Image.open(local_img))

Image 2: This image is named img3.png and contains clouds

True Positives & True Negatives¶

In [9]:
for i, img in df.iterrows():
    if (img.prediction > 0.5) == img.label:
        print(f"Image {i}: {img.description}")
        local_img = ctx.file_access.get_random_local_path()
        ctx.file_access.get(img.remote_source, local_img)
        display(Image.open(local_img))

Image 0: This image is named img1.png and contains hydrocarbon plumes

Image 1: This image is named img2.png and contains hydrocarbon plumes